Skip to content

Formats and timing

Every module has its own starter code, so you can begin at any module. Times include a few minutes of slack, but every room is different. If you’re running behind, the first thing to cut is the Try this section at the end of each module.

Option A: 2-hour meetup, “Build an agent from scratch”

Section titled “Option A: 2-hour meetup, “Build an agent from scratch””

Good for a weeknight meetup or a first event with a new audience.

Time Segment
0:00 Welcome, introduction, setup check (python -m agentlab.check)
0:15 Module 1: Your first model call
0:35 Module 2: Give it tools
1:00 Module 3: The agent loop
1:25 Facilitator demo of Module 5: hijack the agent live (Flag 3 is the best one to show)
1:45 Wrap-up and Q&A

Option B: 2-hour meetup, “Break and secure an AI agent”

Section titled “Option B: 2-hour meetup, “Break and secure an AI agent””

Good for audiences who have built with AI before. It skips the build, so it starts from the Module 3 solution.

Time Segment
0:00 Welcome, setup check
0:10 Walkthrough of the Module 3 agent (read the code together, about 50 lines)
0:25 Module 5: Capture the flags
1:05 Module 6: Defend and prove it
1:45 Wrap-up and Q&A

If you have time to spare, show a short version of Module 4 first so attendees understand the eval runner.

Option C: Half day (about 3.5 hours), “Build and evaluate”

Section titled “Option C: Half day (about 3.5 hours), “Build and evaluate””
Time Segment
0:00 Welcome, introduction, setup check
0:20 Module 1: Your first model call
0:40 Module 2: Give it tools
1:05 Module 3: The agent loop
1:30 Break
1:45 Module 4: Evaluate your agent
2:30 Module 5: Capture the flags (shortened to Flags 1 and 3)
3:05 Wrap-up

Option D: Full day (about 6 hours), the complete workshop

Section titled “Option D: Full day (about 6 hours), the complete workshop”
Time Segment
0:00 Welcome, introductions, setup
0:30 Module 1: Your first model call
0:50 Module 2: Give it tools
1:15 Break
1:30 Module 3: The agent loop
2:00 Module 4: Evaluate your agent
2:45 Lunch
3:30 Module 5: Capture the flags
4:10 Break
4:25 Module 6: Defend and prove it
5:10 Open hacking: add a tool, a test case or a new attack
5:40 Show and tell, wrap-up

No hands-on. The speaker runs everything on stage.

  1. Show the 20-line agent loop from Module 3 and run it (10 minutes)
  2. Run the evals, show a failure, fix it, run them again (10 minutes)
  3. Hijack the agent with the poisoned session, live (10 minutes)
  4. Switch on the defenses and show the security evals go green (10 minutes)
  5. Q&A (5 minutes)

Attendees get the link to do the full workshop at home.